The SMR Business Model Explained: Who Pays, Who Profits, Who Takes the Risk
Nuclear energy is a brilliant technology trapped inside a terrible financing problem, and SMRs are the industry's best attempt to fix that.
Money is the reason most nuclear plants never get built. Not physics. Not public opposition. Not even regulation, though that certainly doesn’t help. The fundamental problem with nuclear energy has always been that the people who benefit from a reactor, over 60 years of operation, are not the people who have to pay for it up front, in a lump sum, before a single electron flows. That mismatch between when money goes in and when it comes back out has killed more nuclear projects than Chernobyl ever did.
Small modular reactors are, among other things, an attempt to fix that mismatch. Whether they actually succeed depends on how the business model gets structured, who absorbs the cost overruns when things go wrong, and whether a new class of buyers, mainly Big Tech companies desperate for always-on clean power, can provide the demand signal the industry has always lacked. This is not a simple story. But it’s one worth understanding, because the business model matters at least as much as the reactor design.
The traditional nuclear money problem
Every energy project has capital costs and operating costs. What makes nuclear unusual is the ratio between them 💡. A natural gas plant is cheap to build and expensive to fuel. A nuclear plant is expensive to build and almost free to operate once it’s running. The fuel costs for nuclear amount to roughly $24 per megawatt-hour in the United States according to data from the Nuclear Energy Institute, a fraction of what gas or coal plants pay.
That sounds like a good deal. The catch is the upfront cost. According to the International Energy Agency’s 2025 estimates, building an SMR in Western countries currently runs about $10,000 per kilowatt of capacity. Build a 100 MW SMR and you’re looking at roughly $1 billion before the reactor has produced a single kilowatt-hour of electricity. For comparison, a standard 1,000 MW conventional nuclear plant can cost $10 billion or more. Lower total bill, same punishing upfront structure.
And here’s what makes lenders nervous 😬:
Nuclear projects have a decades-long history of cost overruns, averaging 117% over initial budget in a study of 180 projects published by academic researchers
Construction timelines routinely run 64% longer than projected
Interest accumulates throughout construction, so delays compound costs exponentially
A reactor that takes 15 years to build at a 5% financing rate ends up dramatically more expensive than a reactor that takes 5 years, even if the overnight cost is higher
The GLOBSEC think tank ran the numbers on exactly this point: an SMR costing $10,000 per kilowatt with a 5-year build can end up cheaper in total than a conventional plant at $6,600 per kilowatt with a 15-year build, once you account for the interest that stacks up during construction. The SMR’s shorter build time does real financial work, even before any economies of series production kick in. But if that 5-year timeline slips to 7 years, the advantage evaporates.
Who actually writes the checks
The Information Technology & Innovation Foundation’s 2025 analysis lists the stakeholders in a typical SMR deal and it reads like the cast of a very stressful ensemble drama 📋:
Vendors: the reactor designers and manufacturers (NuScale, GE Hitachi, Kairos Power, X-energy, Rolls-Royce SMR, and roughly 80 others globally)
Constructors: the EPC firms that actually build the thing on site
Utilities: the traditional owners and operators of power plants
Lenders: banks and institutional investors providing debt financing
Large end users: tech companies, industrial buyers, military customers
Government bodies: national energy departments, loan guarantee programs, grant-making agencies
Ratepayers: the ordinary electricity customers who ultimately pay for regulated utility investments
In the current US model, federal money arrives mainly through three channels. First, R&D grants that help developers get their designs through the regulatory process. Second, tax credits, including a Production Tax Credit set at $30 per megawatt-hour for qualifying nuclear electricity. Third, loan guarantees from the Department of Energy’s Loan Programs Office, which lower borrowing costs by reducing lender risk. The UK has taken a somewhat different approach, committing £210 million in direct grant funding to Rolls-Royce SMR to reduce private investor exposure.
None of this is charity. It’s risk transfer. The government absorbs some of the early-stage uncertainty so that private capital feels safe enough to move. The question is whether the risk transfer is designed well, and the NuScale experience suggests it sometimes isn’t.
The FOAK problem: why the first reactor is always the hardest sell
“FOAK” stands for first-of-a-kind, and it’s the industry’s polite term for “we don’t know what this will actually cost.” Every novel reactor design faces a FOAK build, and FOAK builds are expensive because nothing is optimized yet, supply chains don’t exist at scale, regulators are examining an unfamiliar design, and workers are learning procedures for the first time ⚠️. After the FOAK, subsequent “NOAK” (nth-of-a-kind) builds should be faster and cheaper. The whole SMR business thesis depends on reaching NOAK territory.
NuScale’s Carbon Free Power Project collapse in November 2023 is the FOAK problem made viscerally concrete. The Utah Associated Municipal Power Systems project was supposed to deliver electricity at $55 per megawatt-hour. By cancellation, the projected cost had risen to $89 per megawatt-hour, a 62% increase, driven by inflation, design revisions, and the inherent costs of doing something for the first time. The DOE had committed $1.4 billion in cost-sharing support. None of it was enough.
The Clean Air Task Force’s post-mortem identified several compounding problems:
UAMPS was a collection of small municipal utilities with no experience in nuclear procurement and no capacity to absorb cost overruns on behalf of their ratepayers
The project was structured around a single customer base, so when utilities started dropping out, there was no fallback
NuScale’s pool-based design required large fixed civil works regardless of module count, undermining the “pay for what you need” modularity argument
The project was launched in 2015, well before NRC design certification, meaning customers were committing to a design that hadn’t been approved yet
What this tells us is that the business model failed at least as much as the economics did. The right customer for a FOAK reactor is someone who can absorb cost uncertainty, needs reliable clean power regardless of price, and has a long enough time horizon to wait. Small municipal utilities in competitive electricity markets are essentially the opposite of that customer profile.
Big Tech as the new nuclear buyer
If you need a buyer who has deep pockets, doesn’t flinch at long timelines, needs 24/7 clean power regardless of price, and has a strategic reason to care about energy security, you are describing the AI data center business in 2025 💰. That alignment is not a coincidence. It’s why the past 18 months have produced more nuclear power purchase agreements than the previous decade combined.
Google made history in October 2024 by signing the world’s first corporate SMR power purchase agreement, committing to buy 500 megawatts from Kairos Power’s fleet of molten salt reactors, to be delivered between 2030 and 2035. Amazon followed two days later, anchoring a $500 million investment round in X-energy and committing to 5 gigawatts of SMR capacity by 2039. Microsoft signed a 20-year deal with Constellation Energy to restart Three Mile Island’s Unit 1, delivering 835 megawatts of power. In May 2025, Google went further, committing early-stage capital to Elementl Power for three reactor sites totaling 1.8 gigawatts. Meta issued a request for proposals targeting as much as 4 gigawatts of new nuclear generation.
These are not token investments. Big tech companies signed contracts for more than 10 gigawatts of possible new US nuclear capacity in the past year, according to reporting compiled by Introl. That’s a demand signal the nuclear industry has genuinely never seen from private buyers before.
What makes a Power Purchase Agreement the preferred structure here:
The buyer commits to purchase power at a fixed price for a fixed term, often 20 years, giving lenders the revenue certainty they need to finance construction
The PPA effectively transfers electricity price risk from the reactor owner to the buyer
Tech companies are relatively indifferent to this risk because their electricity costs are small compared to their data center construction and compute costs
Long-term PPAs also help tech companies meet their net-zero commitments with real, always-on generation rather than renewable energy certificates
The honest concern worth naming is that most of these agreements are structured around SMRs that don’t exist yet. Google’s Kairos deal has a first delivery date of 2030. Amazon’s X-energy commitment targets the early 2030s. If those timelines slip, which FOAK history suggests is likely, the tech companies face the choice of absorbing delays or walking away. The PPA structure protects them somewhat, but it doesn’t make the underlying physics of nuclear construction any faster.
The emerging alternatives: own it, lease it, or just buy the steam
The traditional model, where a utility builds, owns, and operates a reactor and sells power on the grid, is only one of several structures now being explored 🔬. The SMR format opens up business models that were simply not practical at gigawatt scale.
The Build-Own-Operate (BOO) model flips the traditional utility structure. Instead of selling the reactor to a customer, the vendor builds it, keeps ownership, and sells electricity or heat as a service. This is how some microreactor developers are positioning themselves, particularly for remote industrial applications. The customer gets clean power without the regulatory burden of owning a nuclear facility. The vendor captures long-term operating revenue. The risk is that the vendor carries all the construction and operational risk on its own balance sheet.
Vendor-financed models are a related variant and, according to market research firm Precedence Research, represent the fastest-growing segment in SMR financing by ownership structure. The vendor finances construction and recoups the cost through long-term service and power contracts. Rolls-Royce SMR’s public positioning leans heavily in this direction: their stated goal is to make their reactors “financeable without the need for government intervention in the long term,” relying on factory-built standardization to make the numbers work for private lenders.
For industrial customers, there’s also a simpler version: just buy the heat. Nuclear reactors produce enormous quantities of process heat, and industries like hydrogen production, aluminum smelting, and semiconductor manufacturing need exactly that kind of continuous, high-temperature energy. An SMR positioned as an industrial heat supplier doesn’t need to touch the electricity grid at all, which sidesteps a range of regulatory complications.
What the EFI Foundation’s 2024 report on SMR bankability makes plain is that no single business model works for all applications. The right structure for powering a remote mine in the Canadian Arctic is completely different from the right structure for supplying a data center campus in Virginia. The industry is still figuring out which models work where, and the next decade of deals will tell us far more than any projection does.
So here’s the question worth actually sitting with: given that FOAK projects in nuclear have historically failed about as often as they’ve succeeded, and given that the entire SMR cost thesis depends on reaching NOAK production volumes, which specific company or country do you think will be the first to deploy a second, third, and fourth identical reactor from the same design — and why?